Intentional looks facilitate faster responding in observers

The study asks whether human eye gaze communicates mental state information (specifically, intentionality) beyond merely signaling direction. While humans readily follow others’ gaze and such gaze-following supports social functions (joint attention, attitude formation, social valuation), it remains unclear if gaze-following is driven solely by directional cues or also by perceived mental states. The authors manipulated gazers’ intentionality by having them either choose their gaze direction (self-chosen; intentional) or follow explicit computer instructions (computer-instructed; non-intentional) and measured how naïve observers perceived and responded to these gaze cues. They hypothesized that if gaze conveys mental states, observers would show faster responses to intentional (self-chosen) gaze even before movement onset, and that gaze-following to targets might be modulated by perceived intentionality, potentially varying with cue-target timing.

Two accounts dominate gaze effects: a directional account (gaze acts like an arrow cue, driving attention via spatial directionality) and a mentalistic account (gaze conveys the gazer’s mental states). Evidence for the directional view shows similar attentional orienting to gaze and non-social directional cues (e.g., arrows). Evidence for the mentalistic view shows reduced gaze-following when observers believe the gazer cannot see the target or when the cue is delivered by a non-social agent (e.g., robots). Prior work on action prediction (Pesquita et al., 2016) found faster identification of targets endogenously chosen by actors versus instructed actions, suggesting sensitivity to control over attention. The present work adopts this logic for eye movements to test whether intentionality information in gaze affects observers’ perception and subsequent responses. The authors also note that eye morphology (high iris–sclera contrast) supports rapid reading of gaze and that social prediction hypotheses posit chosen actions as more predictable from agents’ mental states, implying potential pre-movement communication of intention.

Three preregistered behavioral experiments using naturalistic video stimuli of two naïve female gazers (Gazer 1: 19 years, Asian; Gazer 2: 20 years, White) making left/right gaze shifts under two mental state conditions: self-chosen (intentional) and computer-instructed (non-intentional). Videos were recorded with a 1080p webcam while gazers viewed timing instructions on a monitor. Each clip contained 2000 ms of direct gaze prior to eye movement initiation (T0) and 1000 ms of averted gaze after initiation. Stimuli were presented online via Testable; videos scaled to 1280×720 px at 90% full-screen, color.

• Manipulation of intentionality: On each recorded trial, a double-sided arrow with the word “Choose” prompted self-chosen looks; a directional arrow with “Left/Right” prompted instructed looks. Gazers maintained fixation then looked to a peripheral placeholder dot (eccentricity ~5.72°) for 4000 ms.
• Experiment 1 (prediction of gaze direction before movement): N=81 analyzed undergraduates (70 women, 11 men; mean age 20.48). A priori power analysis targeted d=0.3 with 80% power (n=71). Exclusions: RT<200 ms, RT>1800 ms, or <70% valid trials per participant; 13 excluded. Task: Each trial paused on the last frame before movement (T0); observers indicated upcoming gaze direction (left/right) via keypress. 400 trials across 4 blocks; intentionality and direction intermixed, gazer identity blocked. Analyses: repeated-measures ANOVAs; Greenhouse–Geisser corrections as needed; Bonferroni-corrected paired t-tests.
• Experiment 2 (gaze-cuing with simultaneous targets): N=73 analyzed (61 women, 10 men, 2 other; mean age 20.41). Task: At the first frame after T0, a peripheral target (black asterisk, 1.91°; eccentricity 8.57°, aligned with the eyes) appeared either at the gazed-at side (congruent) or opposite (incongruent); participants localized the target with a keypress. No pause in video; response window 2000 ms. Design: within-subject factors—Gaze intentionality (self-chosen; instructed), Gaze-target congruency (congruent; incongruent), Gazer identity (blocked). 320 trials across 4 blocks (20 reps per condition per block). Exclusions as preregistered.
• Experiment 3 (varying cue–target onset): N=70 analyzed (46 women, 23 men, 1 other; mean age 25.00). Recruitment from student pool and Prolific ($7.31 USD/hour). Same as Experiment 2 but target onset after T0 was 0, 100, 300, or 700 ms (equiprobable). 640 trials across 4 blocks (10 reps per cue-target onset × intentionality × congruency condition per block). Factors: Intentionality, Congruency, Gazer identity (blocked), Cue–target onset.
• Pre-registrations: Exp 1 (OSF: qydwk; 11/14/2022); Exps 2–3 (OSF: mb547; 02/03/2023). Data cleaning and attrition checks were preregistered. Analyses used repeated-measures ANOVAs on accuracy and mean correct RTs.
• Exploratory motion analysis: Optical Flow Analysis quantified motion magnitude within the eye-region ROI in the 2000 ms prior to T0 for each clip and condition. The peak motion frame and ±1 frame defined a temporal window; average motion magnitudes for self-chosen vs instructed were compared with a paired two-tailed t-test (N=120 measurements across sequences).

• Experiment 1 (direction prediction before movement): • Accuracy overall was low (44.46%), with no significant differences by intentionality or direction. Faster RTs for self-chosen vs computer-instructed gaze: main effect of Gaze intentionality, F(1,80)=8.581, p=0.004, ηp²=0.097 (self-chosen faster). Also a main effect of direction favoring left over right, F(1,80)=4.166, p=0.045, ηp²=0.049, and an interaction F(1,80)=7.617, p=0.007, ηp²=0.087, driven by faster responses to right vs left only in computer-instructed trials. Effects did not differ by gazer identity.
• Experiment 2 (gaze-cuing, target at 0 ms after T0): • High accuracy (~97%) with no effects on accuracy. • Classic gaze-following: congruent faster than incongruent, F(1,72)=17.764, p<0.001, ηp²=0.198. • Main effect of intentionality: F(1,72)=6.527, p=0.013, ηp²=0.083, with overall slower RTs for self-chosen vs instructed gaze at 0 ms onset; no interaction between intentionality and congruency (F(1,72)=0.796, p=0.375), and no moderation by gazer identity of the congruency effect.
• Experiment 3 (gaze-cuing with variable cue–target onsets): • Accuracy: main effect of cue–target onset (better at 0 and 300 ms vs 700 ms); no other accuracy effects. • RTs: robust gaze-following (congruent faster), F(1,69)=51.355, p<0.001, ηp²=0.427; strong foreperiod effect (longer SOA faster), F(2.012,138.801)=158.230, p<0.001, ηp²=0.696. • Gaze-following was time-dependent (facilitation at 100 and 300 ms): onset×congruency interaction F(2.691,185.661)=3.966, p=0.012, ηp²=0.054. • Intentionality main effect: faster responses to self-chosen vs instructed, F(1,69)=31.817, p<0.001, ηp²=0.316; qualified by onset×intentionality interaction, F(3,207)=7.704, p<0.001, ηp²=0.100: self-chosen faster at 100, 300, and 700 ms; at 0 ms, self-chosen slower (replicating Exp 2).
• Exploratory motion analysis (optical flow): greater motion magnitude in eye region for self-chosen (M=1.24, SD=0.14) vs computer-instructed (M=0.72, SD=0.19); t(2)=6.882, p=0.020, d=3.973, indicating subtle kinematic differences preceding movement that may signal intentionality. Overall: Observers were sensitive to intentionality in gaze. Intentional looks facilitated faster responding, particularly when given time to process the cue, while the magnitude of the gaze-following effect itself did not reliably differ by intentionality.

Findings indicate that human observers rapidly and implicitly extract mental state information (intentionality) from gaze cues, leading to faster responses for intentional (self-chosen) looks even before movement onset (Exp 1) and at longer cue–target intervals (Exp 3). At the same time, classic direction-based gaze-following (congruent vs incongruent targets) was robust but not modulated by intentionality across Experiments 2 and 3, suggesting that mentalistic and directional components of gaze may operate in parallel on different time scales. The pattern—early broad response facilitation without accuracy advantages, time-dependent facilitation for intentional gaze, and independent gaze-following—supports an implicit mentalizing mechanism that precedes and runs alongside more deliberate directionality-based orienting. Individual variability in gazers’ effectiveness communicating mental states was observed, but did not alter the central conclusions. Exploratory kinematic analysis suggests subtle pre-movement motion within the eye region as a candidate signal through which intentionality is conveyed.

The study demonstrates that human gaze communicates mental states: intentional (self-chosen) looks facilitate observers’ responses, likely via subtle motion cues within the eye region. Directional (gaze-following) and mentalistic influences appear to be processed in parallel, on distinct time courses. Future research should: expand the diversity and number of gazers; examine effects of gazer and observer gender; test how modality of instruction (e.g., auditory) or task emphasis (timing vs location) shapes mental state perception; manipulate or remove critical motion segments to identify causal kinematic features; and use high-speed eye tracking to precisely characterize motion parameters (e.g., velocity, trajectory) underpinning mental state communication.

• Stimuli featured only two gazers, limiting generalizability across individual differences in signaling effectiveness.
• Sample demographics were largely female observers; potential interactions between gazer and participant gender were not examined and warrant study.
• Laboratory-like control may not capture complex real-world contexts; contextual dynamics (e.g., natural interaction patterns, eye contact frequency) could modulate mental state extraction from gaze.
• The intentionality manipulation relied on instruction context; generalization to other forms of intentional control should be tested.